WinRAR 4.20 gets multi-threaded support

[ninaholic37]

When you guys are running these compression tests, how do you know the performance improvement isn't being helped by the files being cached in RAM from your previous test?

Shut down / turn back on your computer after each test ?

I can run a 7zip compression test when I get home.

Any ideas on what you want in the compression test is a source folder with just mp3's ok?

For a mix, I tried compressing/decompressing my "Program Files" folder:

Program Files.rar 8:45/1:39 550MB (WinRar 4.20 default / multi-thread)
Program Files.7z 14:17/2:29 432MB (7-zip 9.20 default - 2 threads)
Program Files.7z 10:26/2:08 434MB (7-zip 9.28 Alpha default - 4 threads)

This was on 32-bit. 64-bit and 4C/8T may give it different speed ratios...

One thing I found odd, is that just opening the RAR file (to browse) seems to take around 10 seconds, but opening the 7z file takes 0 seconds. Maybe this "initial RAR loading" is to help save time when you're trying to extract 'individual files' (?), as 7-zip seems to "filter through a chunk" to get to a single file or something. This difference makes me want to read up more on how these RAR/7z files work internally, seems kind of intriguing, this.

 

[MentalIlness]

CPU @ 4.0Ghz

Nothing disabled

Core 5/6/7/8 Disabled

Core 3/4/7/8 Disabled

Core 7/8 Disabled

 

[ShintaiDK]

Interesting. Im also around 80-90% Load when running the test. Wonder why it cant reach 100%.

Also rerunning ti quickly etc can really screw results. My results are up to 20% different from one another.

 

[billyb0b]

very nice... i've always loved winrar and stayed loyal to it for about 10 years. i recently switched over to 7zip because it's free. the interface is simple and clean. and the 7z archive was packing a little better than winrar was with rar

 

[MentalIlness]

Would a SSD make any difference in this ?

 

[boxleitnerb]

i7-2600K@4GHz (all threads)
7,500kB/s, CPU load 70-90%

 

[Makaveli]

Shut down / turn back on your computer after each test ?


For a mix, I tried compressing/decompressing my "Program Files" folder:

Program Files.rar 8:45/1:39 550MB (WinRar 4.20 default / multi-thread)
Program Files.7z 14:17/2:29 432MB (7-zip 9.20 default - 2 threads)
Program Files.7z 10:26/2:08 434MB (7-zip 9.28 Alpha default - 4 threads)

This was on 32-bit. 64-bit and 4C/8T may give it different speed ratios...

One thing I found odd, is that just opening the RAR file (to browse) seems to take around 10 seconds, but opening the 7z file takes 0 seconds. Maybe this "initial RAR loading" is to help save time when you're trying to extract 'individual files' (?), as 7-zip seems to "filter through a chunk" to get to a single file or something. This difference makes me want to read up more on how these RAR/7z files work internally, seems kind of intriguing, this.

Interesting find so based on your results 7 zip is a abit slower but it seemed to make the file smaller. Haven't noticed the browsing difference will check this out now.

Would a SSD make any difference in this ?

I was wondering this myself so I will do a test with source data on my SSD aswell as my RAM disc.

i7-2600K@4GHz (all threads)
7,500kB/s, CPU load 70-90%

Based on your results looks like SB is just a tad faster than Bulldozer when both are at 4Ghz on this test. Which makes sense cause BD is able to close the gap when there is an app that can actually use alot of cores.

Can we get some IVY/SB-E results at 4.0ghz aswell?

 

[AtenRa]

It seams SSD is having a big role in the performance,

Intel Core i7 920@4GHz with Intel SSD 320 120GB SATA-II

WinRar 4.2 64bit

Edit: Perhaps its the Triple Channel Memory ??

 

[Makaveli]

It seams SSD is having a big role in the performance,

Intel Core i7 920@4GHz with Intel SSD 320 120GB SATA-II

WinRar 4.2 64bit

Edit: Perhaps its the Triple Channel Memory ??

It is interesting that your numbers are higher than SB at 4ghz.

I think you maybe right about the extra memory bandwidth.

I need someone with SB-E @ 4ghz which is a quad channel setup to post if there numbers are higher than what i posted this should confirm your theory.

 

[AtenRa]

Ill run with dual channel in a few minutes and see if ill get lower results.

 

[soccerballtux]

I don't have it. People with FX processors need to come forward

Isn't just you. Nobody has it. LOL.

 

[soccerballtux]

...

Awesome thanks!
Can you re-run the first one? It says current is ~7400, but result was ~7000.

Bulldozer modules mapped 1/2 + 3/4 + 5/6 + 7/8...I wonder what would happen if we tried one with 2/4/6/8 disabled. Interesting that the per-module score jumped significantly when you disabled entire modules. Perhaps the modules are a little L3 cache or L3 bandwidth starved? Edit: probably bandwidth...

 

[soccerballtux]

i7-2600K@4GHz (all threads)
7,500kB/s, CPU load 70-90%

Wow, performance parity with Bulldozer clock for clock!

 

[AtenRa]

OK, final verdict ...... RAM Bandwidth

Triple Channel 1532MHz 9-9-9-24

Dual Channel 1532MHz 9-9-9-24

Triple Channel 1150MHz 9-9-9-24

Dual Channel 1150MHz 9-9-9-24

Socket 2011 owners favorable benchmark

 

[soccerballtux]

^nice.

So if we overclocked a Bulldozer to 4.8Ghz, we would get over 9000 and be faster than a 4Ghz SB-E...

 

[Makaveli]

OK, final verdict ...... RAM Bandwidth

Triple Channel 1532MHz 9-9-9-24

Dual Channel 1532MHz 9-9-9-24

Triple Channel 1150MHz 9-9-9-24

Dual Channel 1150MHz 9-9-9-24

Socket 2011 owners favorable benchmark

Awesome thanks for the testing. So it looks like winrar cares more about core count and memory bandwidth over IPC.

 

[Makaveli]

^nice.

So if we overclocked a Bulldozer to 4.8Ghz, we would get over 9000 and be faster than a 4Ghz SB-E...

hmm no!

I'm doing 12k at 4Ghz, SB-E should best my results do to having one extra memory channel.

BD would have to be at 5.2Ghz+ to come close!

 

[AtenRa]

Depends on the ram he was running with the FX.

@MentalIlness
what frequency do you run the ram with the BD ??

 

[ShintaiDK]

Thats one crazy memory dependency. Specially compared to 7z.
I wonder if it will be improved in the future. WinRAR just seems overly dependent on main memory. Atleast for the inbuildt bench.

 

[boxleitnerb]

Wow, performance parity with Bulldozer clock for clock!

I only have puny DDR3-1333, so the results could be better.

 

[Makaveli]

I only have puny DDR3-1333, so the results could be better.

That would explain it based on everyone results so far the primary factors for this benchmark is Core count, Memory bandwidth and Memory speed!

 

[Rhoxed]

1090T @ 4ghz, 2.9ghz NB, ram @ 1760 8-8-8-21 = 7,068

 

[soccerballtux]

1090T @ 4ghz, 2.9ghz NB, ram @ 1760 8-8-8-21 = 7,068

so there's only ever so slightly a reason to get bulldozer...

 

[soccerballtux]

hmm no!

I'm doing 12k at 4Ghz, SB-E should best my results do to having one extra memory channel.

BD would have to be at 5.2Ghz+ to come close!

woops, Nehalem.
Looks like we can meet the Nehalems. About time.

 

[Magic Carpet]

1090T @ 4ghz, 2.9ghz NB, ram @ 1760 8-8-8-21 = 7,068

Can you bench your Thuban @ ~3,250 mhz? and memory @ 1600 7-7-7-24, single channel?

Thanks.